xref: /freebsd/sys/dev/cxgbe/adapter.h (revision ba3c1f5972d7b90feb6e6da47905ff2757e0fe57)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: Navdeep Parhar <np@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * $FreeBSD$
30  *
31  */
32 
33 #ifndef __T4_ADAPTER_H__
34 #define __T4_ADAPTER_H__
35 
36 #include <sys/kernel.h>
37 #include <sys/bus.h>
38 #include <sys/counter.h>
39 #include <sys/rman.h>
40 #include <sys/types.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/rwlock.h>
44 #include <sys/seqc.h>
45 #include <sys/sx.h>
46 #include <sys/vmem.h>
47 #include <vm/uma.h>
48 
49 #include <dev/pci/pcivar.h>
50 #include <dev/pci/pcireg.h>
51 #include <machine/bus.h>
52 #include <sys/socket.h>
53 #include <sys/sysctl.h>
54 #include <sys/taskqueue.h>
55 #include <net/ethernet.h>
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_media.h>
59 #include <net/pfil.h>
60 #include <netinet/in.h>
61 #include <netinet/tcp_lro.h>
62 
63 #include "offload.h"
64 #include "t4_ioctl.h"
65 #include "common/t4_msg.h"
66 #include "firmware/t4fw_interface.h"
67 
68 #define KTR_CXGBE	KTR_SPARE3
69 MALLOC_DECLARE(M_CXGBE);
70 #define CXGBE_UNIMPLEMENTED(s) \
71     panic("%s (%s, line %d) not implemented yet.", s, __FILE__, __LINE__)
72 
73 /*
74  * Same as LIST_HEAD from queue.h.  This is to avoid conflict with LinuxKPI's
75  * LIST_HEAD when building iw_cxgbe.
76  */
77 #define	CXGBE_LIST_HEAD(name, type)					\
78 struct name {								\
79 	struct type *lh_first;	/* first element */			\
80 }
81 
82 #ifndef SYSCTL_ADD_UQUAD
83 #define SYSCTL_ADD_UQUAD SYSCTL_ADD_QUAD
84 #define sysctl_handle_64 sysctl_handle_quad
85 #define CTLTYPE_U64 CTLTYPE_QUAD
86 #endif
87 
88 SYSCTL_DECL(_hw_cxgbe);
89 
90 struct adapter;
91 typedef struct adapter adapter_t;
92 
93 enum {
94 	/*
95 	 * All ingress queues use this entry size.  Note that the firmware event
96 	 * queue and any iq expecting CPL_RX_PKT in the descriptor needs this to
97 	 * be at least 64.
98 	 */
99 	IQ_ESIZE = 64,
100 
101 	/* Default queue sizes for all kinds of ingress queues */
102 	FW_IQ_QSIZE = 256,
103 	RX_IQ_QSIZE = 1024,
104 
105 	/* All egress queues use this entry size */
106 	EQ_ESIZE = 64,
107 
108 	/* Default queue sizes for all kinds of egress queues */
109 	CTRL_EQ_QSIZE = 1024,
110 	TX_EQ_QSIZE = 1024,
111 
112 	SW_ZONE_SIZES = 4,	/* cluster, jumbop, jumbo9k, jumbo16k */
113 	CL_METADATA_SIZE = CACHE_LINE_SIZE,
114 
115 	SGE_MAX_WR_NDESC = SGE_MAX_WR_LEN / EQ_ESIZE, /* max WR size in desc */
116 	TX_SGL_SEGS = 39,
117 	TX_SGL_SEGS_TSO = 38,
118 	TX_SGL_SEGS_VM = 38,
119 	TX_SGL_SEGS_VM_TSO = 37,
120 	TX_SGL_SEGS_EO_TSO = 30,	/* XXX: lower for IPv6. */
121 	TX_SGL_SEGS_VXLAN_TSO = 37,
122 	TX_WR_FLITS = SGE_MAX_WR_LEN / 8
123 };
124 
125 enum {
126 	/* adapter intr_type */
127 	INTR_INTX	= (1 << 0),
128 	INTR_MSI	= (1 << 1),
129 	INTR_MSIX	= (1 << 2)
130 };
131 
132 enum {
133 	XGMAC_MTU	= (1 << 0),
134 	XGMAC_PROMISC	= (1 << 1),
135 	XGMAC_ALLMULTI	= (1 << 2),
136 	XGMAC_VLANEX	= (1 << 3),
137 	XGMAC_UCADDR	= (1 << 4),
138 	XGMAC_MCADDRS	= (1 << 5),
139 
140 	XGMAC_ALL	= 0xffff
141 };
142 
143 enum {
144 	/* flags understood by begin_synchronized_op */
145 	HOLD_LOCK	= (1 << 0),
146 	SLEEP_OK	= (1 << 1),
147 	INTR_OK		= (1 << 2),
148 
149 	/* flags understood by end_synchronized_op */
150 	LOCK_HELD	= HOLD_LOCK,
151 };
152 
153 enum {
154 	/* adapter flags.  synch_op or adapter_lock. */
155 	FULL_INIT_DONE	= (1 << 0),
156 	FW_OK		= (1 << 1),
157 	CHK_MBOX_ACCESS	= (1 << 2),
158 	MASTER_PF	= (1 << 3),
159 	BUF_PACKING_OK	= (1 << 6),
160 	IS_VF		= (1 << 7),
161 	KERN_TLS_ON	= (1 << 8),	/* HW is configured for KERN_TLS */
162 	CXGBE_BUSY	= (1 << 9),
163 
164 	/* adapter error_flags.  reg_lock for HW_OFF_LIMITS, atomics for the rest. */
165 	ADAP_STOPPED	= (1 << 0),	/* Adapter has been stopped. */
166 	ADAP_FATAL_ERR	= (1 << 1),	/* Encountered a fatal error. */
167 	HW_OFF_LIMITS	= (1 << 2),	/* off limits to all except reset_thread */
168 	ADAP_CIM_ERR	= (1 << 3),	/* Error was related to FW/CIM. */
169 
170 	/* port flags */
171 	HAS_TRACEQ	= (1 << 3),
172 	FIXED_IFMEDIA	= (1 << 4),	/* ifmedia list doesn't change. */
173 
174 	/* VI flags */
175 	DOOMED		= (1 << 0),
176 	VI_INIT_DONE	= (1 << 1),
177 	/* 1 << 2 is unused, was VI_SYSCTL_CTX */
178 	TX_USES_VM_WR	= (1 << 3),
179 	VI_SKIP_STATS	= (1 << 4),
180 
181 	/* adapter debug_flags */
182 	DF_DUMP_MBOX		= (1 << 0),	/* Log all mbox cmd/rpl. */
183 	DF_LOAD_FW_ANYTIME	= (1 << 1),	/* Allow LOAD_FW after init */
184 	DF_DISABLE_TCB_CACHE	= (1 << 2),	/* Disable TCB cache (T6+) */
185 	DF_DISABLE_CFG_RETRY	= (1 << 3),	/* Disable fallback config */
186 	DF_VERBOSE_SLOWINTR	= (1 << 4),	/* Chatty slow intr handler */
187 };
188 
189 #define IS_DOOMED(vi)	((vi)->flags & DOOMED)
190 #define SET_DOOMED(vi)	do {(vi)->flags |= DOOMED;} while (0)
191 #define IS_BUSY(sc)	((sc)->flags & CXGBE_BUSY)
192 #define SET_BUSY(sc)	do {(sc)->flags |= CXGBE_BUSY;} while (0)
193 #define CLR_BUSY(sc)	do {(sc)->flags &= ~CXGBE_BUSY;} while (0)
194 
195 struct vi_info {
196 	device_t dev;
197 	struct port_info *pi;
198 	struct adapter *adapter;
199 
200 	if_t ifp;
201 	struct pfil_head *pfil;
202 
203 	unsigned long flags;
204 	int if_flags;
205 
206 	uint16_t *rss, *nm_rss;
207 	uint16_t viid;		/* opaque VI identifier */
208 	uint16_t smt_idx;
209 	uint16_t vin;
210 	uint8_t vfvld;
211 	int16_t  xact_addr_filt;/* index of exact MAC address filter */
212 	uint16_t rss_size;	/* size of VI's RSS table slice */
213 	uint16_t rss_base;	/* start of VI's RSS table slice */
214 	int hashen;
215 
216 	int nintr;
217 	int first_intr;
218 
219 	/* These need to be int as they are used in sysctl */
220 	int ntxq;		/* # of tx queues */
221 	int first_txq;		/* index of first tx queue */
222 	int rsrv_noflowq;	/* Reserve queue 0 for non-flowid packets */
223 	int nrxq;		/* # of rx queues */
224 	int first_rxq;		/* index of first rx queue */
225 	int nofldtxq;		/* # of offload tx queues */
226 	int first_ofld_txq;	/* index of first offload tx queue */
227 	int nofldrxq;		/* # of offload rx queues */
228 	int first_ofld_rxq;	/* index of first offload rx queue */
229 	int nnmtxq;
230 	int first_nm_txq;
231 	int nnmrxq;
232 	int first_nm_rxq;
233 	int tmr_idx;
234 	int ofld_tmr_idx;
235 	int pktc_idx;
236 	int ofld_pktc_idx;
237 	int qsize_rxq;
238 	int qsize_txq;
239 
240 	struct timeval last_refreshed;
241 	struct fw_vi_stats_vf stats;
242 	struct mtx tick_mtx;
243 	struct callout tick;
244 
245 	struct sysctl_ctx_list ctx;
246 	struct sysctl_oid *rxq_oid;
247 	struct sysctl_oid *txq_oid;
248 	struct sysctl_oid *nm_rxq_oid;
249 	struct sysctl_oid *nm_txq_oid;
250 	struct sysctl_oid *ofld_rxq_oid;
251 	struct sysctl_oid *ofld_txq_oid;
252 
253 	uint8_t hw_addr[ETHER_ADDR_LEN]; /* factory MAC address, won't change */
254 	u_int txq_rr;
255 	u_int rxq_rr;
256 };
257 
258 struct tx_ch_rl_params {
259 	enum fw_sched_params_rate ratemode;	/* %port (REL) or kbps (ABS) */
260 	uint32_t maxrate;
261 };
262 
263 /* CLRL state */
264 enum clrl_state {
265 	CS_UNINITIALIZED = 0,
266 	CS_PARAMS_SET,			/* sw parameters have been set. */
267 	CS_HW_UPDATE_REQUESTED,		/* async HW update requested. */
268 	CS_HW_UPDATE_IN_PROGRESS,	/* sync hw update in progress. */
269 	CS_HW_CONFIGURED		/* configured in the hardware. */
270 };
271 
272 /* CLRL flags */
273 enum {
274 	CF_USER		= (1 << 0),	/* was configured by driver ioctl. */
275 };
276 
277 struct tx_cl_rl_params {
278 	enum clrl_state state;
279 	int refcount;
280 	uint8_t flags;
281 	enum fw_sched_params_rate ratemode;	/* %port REL or ABS value */
282 	enum fw_sched_params_unit rateunit;	/* kbps or pps (when ABS) */
283 	enum fw_sched_params_mode mode;		/* aggr or per-flow */
284 	uint32_t maxrate;
285 	uint16_t pktsize;
286 	uint16_t burstsize;
287 };
288 
289 /* Tx scheduler parameters for a channel/port */
290 struct tx_sched_params {
291 	/* Channel Rate Limiter */
292 	struct tx_ch_rl_params ch_rl;
293 
294 	/* Class WRR */
295 	/* XXX */
296 
297 	/* Class Rate Limiter (including the default pktsize and burstsize). */
298 	int pktsize;
299 	int burstsize;
300 	struct tx_cl_rl_params cl_rl[];
301 };
302 
303 struct port_info {
304 	device_t dev;
305 	struct adapter *adapter;
306 
307 	struct vi_info *vi;
308 	int nvi;
309 	int up_vis;
310 	int uld_vis;
311 	bool vxlan_tcam_entry;
312 
313 	struct tx_sched_params *sched_params;
314 
315 	struct mtx pi_lock;
316 	char lockname[16];
317 	unsigned long flags;
318 
319 	uint8_t  lport;		/* associated offload logical port */
320 	int8_t   mdio_addr;
321 	uint8_t  port_type;
322 	uint8_t  mod_type;
323 	uint8_t  port_id;
324 	uint8_t  tx_chan;
325 	uint8_t  mps_bg_map;	/* rx MPS buffer group bitmap */
326 	uint8_t  rx_e_chan_map;	/* rx TP e-channel bitmap */
327 	uint8_t  rx_c_chan;	/* rx TP c-channel */
328 
329 	struct link_config link_cfg;
330 	struct ifmedia media;
331 
332 	struct port_stats stats;
333 	u_int tnl_cong_drops;
334 	u_int tx_parse_error;
335 	int fcs_reg;
336 	uint64_t fcs_base;
337 
338 	struct sysctl_ctx_list ctx;
339 };
340 
341 #define	IS_MAIN_VI(vi)		((vi) == &((vi)->pi->vi[0]))
342 
343 struct cluster_metadata {
344 	uma_zone_t zone;
345 	caddr_t cl;
346 	u_int refcount;
347 };
348 
349 struct fl_sdesc {
350 	caddr_t cl;
351 	uint16_t nmbuf;	/* # of driver originated mbufs with ref on cluster */
352 	int16_t moff;	/* offset of metadata from cl */
353 	uint8_t zidx;
354 };
355 
356 struct tx_desc {
357 	__be64 flit[8];
358 };
359 
360 struct tx_sdesc {
361 	struct mbuf *m;		/* m_nextpkt linked chain of frames */
362 	uint8_t desc_used;	/* # of hardware descriptors used by the WR */
363 };
364 
365 
366 #define IQ_PAD (IQ_ESIZE - sizeof(struct rsp_ctrl) - sizeof(struct rss_header))
367 struct iq_desc {
368 	struct rss_header rss;
369 	uint8_t cpl[IQ_PAD];
370 	struct rsp_ctrl rsp;
371 };
372 #undef IQ_PAD
373 CTASSERT(sizeof(struct iq_desc) == IQ_ESIZE);
374 
375 enum {
376 	/* iq type */
377 	IQ_OTHER	= FW_IQ_IQTYPE_OTHER,
378 	IQ_ETH		= FW_IQ_IQTYPE_NIC,
379 	IQ_OFLD		= FW_IQ_IQTYPE_OFLD,
380 
381 	/* iq flags */
382 	IQ_SW_ALLOCATED	= (1 << 0),	/* sw resources allocated */
383 	IQ_HAS_FL	= (1 << 1),	/* iq associated with a freelist */
384 	IQ_RX_TIMESTAMP	= (1 << 2),	/* provide the SGE rx timestamp */
385 	IQ_LRO_ENABLED	= (1 << 3),	/* iq is an eth rxq with LRO enabled */
386 	IQ_ADJ_CREDIT	= (1 << 4),	/* hw is off by 1 credit for this iq */
387 	IQ_HW_ALLOCATED	= (1 << 5),	/* fw/hw resources allocated */
388 
389 	/* iq state */
390 	IQS_DISABLED	= 0,
391 	IQS_BUSY	= 1,
392 	IQS_IDLE	= 2,
393 
394 	/* netmap related flags */
395 	NM_OFF	= 0,
396 	NM_ON	= 1,
397 	NM_BUSY	= 2,
398 };
399 
400 enum {
401 	CPL_COOKIE_RESERVED = 0,
402 	CPL_COOKIE_FILTER,
403 	CPL_COOKIE_DDP0,
404 	CPL_COOKIE_DDP1,
405 	CPL_COOKIE_TOM,
406 	CPL_COOKIE_HASHFILTER,
407 	CPL_COOKIE_ETHOFLD,
408 	CPL_COOKIE_KERN_TLS,
409 
410 	NUM_CPL_COOKIES = 8	/* Limited by M_COOKIE.  Do not increase. */
411 };
412 
413 struct sge_iq;
414 struct rss_header;
415 typedef int (*cpl_handler_t)(struct sge_iq *, const struct rss_header *,
416     struct mbuf *);
417 typedef int (*an_handler_t)(struct sge_iq *, const struct rsp_ctrl *);
418 typedef int (*fw_msg_handler_t)(struct adapter *, const __be64 *);
419 
420 /*
421  * Ingress Queue: T4 is producer, driver is consumer.
422  */
423 struct sge_iq {
424 	uint16_t flags;
425 	uint8_t qtype;
426 	volatile int state;
427 	struct adapter *adapter;
428 	struct iq_desc  *desc;	/* KVA of descriptor ring */
429 	int8_t   intr_pktc_idx;	/* packet count threshold index */
430 	uint8_t  gen;		/* generation bit */
431 	uint8_t  intr_params;	/* interrupt holdoff parameters */
432 	int8_t   cong_drop;	/* congestion drop settings for the queue */
433 	uint16_t qsize;		/* size (# of entries) of the queue */
434 	uint16_t sidx;		/* index of the entry with the status page */
435 	uint16_t cidx;		/* consumer index */
436 	uint16_t cntxt_id;	/* SGE context id for the iq */
437 	uint16_t abs_id;	/* absolute SGE id for the iq */
438 	int16_t intr_idx;	/* interrupt used by the queue */
439 
440 	STAILQ_ENTRY(sge_iq) link;
441 
442 	bus_dma_tag_t desc_tag;
443 	bus_dmamap_t desc_map;
444 	bus_addr_t ba;		/* bus address of descriptor ring */
445 };
446 
447 enum {
448 	/* eq type */
449 	EQ_CTRL		= 1,
450 	EQ_ETH		= 2,
451 	EQ_OFLD		= 3,
452 
453 	/* eq flags */
454 	EQ_SW_ALLOCATED	= (1 << 0),	/* sw resources allocated */
455 	EQ_HW_ALLOCATED	= (1 << 1),	/* hw/fw resources allocated */
456 	EQ_ENABLED	= (1 << 3),	/* open for business */
457 	EQ_QFLUSH	= (1 << 4),	/* if_qflush in progress */
458 };
459 
460 /* Listed in order of preference.  Update t4_sysctls too if you change these */
461 enum {DOORBELL_UDB, DOORBELL_WCWR, DOORBELL_UDBWC, DOORBELL_KDB};
462 
463 /*
464  * Egress Queue: driver is producer, T4 is consumer.
465  *
466  * Note: A free list is an egress queue (driver produces the buffers and T4
467  * consumes them) but it's special enough to have its own struct (see sge_fl).
468  */
469 struct sge_eq {
470 	unsigned int flags;	/* MUST be first */
471 	unsigned int cntxt_id;	/* SGE context id for the eq */
472 	unsigned int abs_id;	/* absolute SGE id for the eq */
473 	uint8_t type;		/* EQ_CTRL/EQ_ETH/EQ_OFLD */
474 	uint8_t doorbells;
475 	uint8_t tx_chan;	/* tx channel used by the eq */
476 	struct mtx eq_lock;
477 
478 	struct tx_desc *desc;	/* KVA of descriptor ring */
479 	volatile uint32_t *udb;	/* KVA of doorbell (lies within BAR2) */
480 	u_int udb_qid;		/* relative qid within the doorbell page */
481 	uint16_t sidx;		/* index of the entry with the status page */
482 	uint16_t cidx;		/* consumer idx (desc idx) */
483 	uint16_t pidx;		/* producer idx (desc idx) */
484 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
485 	uint16_t dbidx;		/* pidx of the most recent doorbell */
486 	uint16_t iqid;		/* cached iq->cntxt_id (see iq below) */
487 	volatile u_int equiq;	/* EQUIQ outstanding */
488 	struct sge_iq *iq;	/* iq that receives egr_update for the eq */
489 
490 	bus_dma_tag_t desc_tag;
491 	bus_dmamap_t desc_map;
492 	bus_addr_t ba;		/* bus address of descriptor ring */
493 	char lockname[16];
494 };
495 
496 struct rx_buf_info {
497 	uma_zone_t zone;	/* zone that this cluster comes from */
498 	uint16_t size1;		/* same as size of cluster: 2K/4K/9K/16K.
499 				 * hwsize[hwidx1] = size1.  No spare. */
500 	uint16_t size2;		/* hwsize[hwidx2] = size2.
501 				 * spare in cluster = size1 - size2. */
502 	int8_t hwidx1;		/* SGE bufsize idx for size1 */
503 	int8_t hwidx2;		/* SGE bufsize idx for size2 */
504 	uint8_t type;		/* EXT_xxx type of the cluster */
505 };
506 
507 enum {
508 	NUM_MEMWIN = 3,
509 
510 	MEMWIN0_APERTURE = 2048,
511 	MEMWIN0_BASE     = 0x1b800,
512 
513 	MEMWIN1_APERTURE = 32768,
514 	MEMWIN1_BASE     = 0x28000,
515 
516 	MEMWIN2_APERTURE_T4 = 65536,
517 	MEMWIN2_BASE_T4     = 0x30000,
518 
519 	MEMWIN2_APERTURE_T5 = 128 * 1024,
520 	MEMWIN2_BASE_T5     = 0x60000,
521 };
522 
523 struct memwin {
524 	struct rwlock mw_lock __aligned(CACHE_LINE_SIZE);
525 	uint32_t mw_base;	/* constant after setup_memwin */
526 	uint32_t mw_aperture;	/* ditto */
527 	uint32_t mw_curpos;	/* protected by mw_lock */
528 };
529 
530 enum {
531 	FL_STARVING	= (1 << 0), /* on the adapter's list of starving fl's */
532 	FL_DOOMED	= (1 << 1), /* about to be destroyed */
533 	FL_BUF_PACKING	= (1 << 2), /* buffer packing enabled */
534 	FL_BUF_RESUME	= (1 << 3), /* resume from the middle of the frame */
535 };
536 
537 #define FL_RUNNING_LOW(fl) \
538     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) <= fl->lowat)
539 #define FL_NOT_RUNNING_LOW(fl) \
540     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) >= 2 * fl->lowat)
541 
542 struct sge_fl {
543 	struct mtx fl_lock;
544 	__be64 *desc;		/* KVA of descriptor ring, ptr to addresses */
545 	struct fl_sdesc *sdesc;	/* KVA of software descriptor ring */
546 	uint16_t zidx;		/* refill zone idx */
547 	uint16_t safe_zidx;
548 	uint16_t lowat;		/* # of buffers <= this means fl needs help */
549 	int flags;
550 	uint16_t buf_boundary;
551 
552 	/* The 16b idx all deal with hw descriptors */
553 	uint16_t dbidx;		/* hw pidx after last doorbell */
554 	uint16_t sidx;		/* index of status page */
555 	volatile uint16_t hw_cidx;
556 
557 	/* The 32b idx are all buffer idx, not hardware descriptor idx */
558 	uint32_t cidx;		/* consumer index */
559 	uint32_t pidx;		/* producer index */
560 
561 	uint32_t dbval;
562 	u_int rx_offset;	/* offset in fl buf (when buffer packing) */
563 	volatile uint32_t *udb;
564 
565 	uint64_t cl_allocated;	/* # of clusters allocated */
566 	uint64_t cl_recycled;	/* # of clusters recycled */
567 	uint64_t cl_fast_recycled; /* # of clusters recycled (fast) */
568 
569 	/* These 3 are valid when FL_BUF_RESUME is set, stale otherwise. */
570 	struct mbuf *m0;
571 	struct mbuf **pnext;
572 	u_int remaining;
573 
574 	uint16_t qsize;		/* # of hw descriptors (status page included) */
575 	uint16_t cntxt_id;	/* SGE context id for the freelist */
576 	TAILQ_ENTRY(sge_fl) link; /* All starving freelists */
577 	bus_dma_tag_t desc_tag;
578 	bus_dmamap_t desc_map;
579 	char lockname[16];
580 	bus_addr_t ba;		/* bus address of descriptor ring */
581 };
582 
583 struct mp_ring;
584 
585 struct txpkts {
586 	uint8_t wr_type;	/* type 0 or type 1 */
587 	uint8_t npkt;		/* # of packets in this work request */
588 	uint8_t len16;		/* # of 16B pieces used by this work request */
589 	uint8_t score;
590 	uint8_t max_npkt;	/* maximum number of packets allowed */
591 	uint16_t plen;		/* total payload (sum of all packets) */
592 
593 	/* straight from fw_eth_tx_pkts_vm_wr. */
594 	__u8   ethmacdst[6];
595 	__u8   ethmacsrc[6];
596 	__be16 ethtype;
597 	__be16 vlantci;
598 
599 	struct mbuf *mb[15];
600 };
601 
602 /* txq: SGE egress queue + what's needed for Ethernet NIC */
603 struct sge_txq {
604 	struct sge_eq eq;	/* MUST be first */
605 
606 	if_t ifp;		/* the interface this txq belongs to */
607 	struct mp_ring *r;	/* tx software ring */
608 	struct tx_sdesc *sdesc;	/* KVA of software descriptor ring */
609 	struct sglist *gl;
610 	__be32 cpl_ctrl0;	/* for convenience */
611 	int tc_idx;		/* traffic class */
612 	uint64_t last_tx;	/* cycle count when eth_tx was last called */
613 	struct txpkts txp;
614 
615 	struct task tx_reclaim_task;
616 	/* stats for common events first */
617 
618 	uint64_t txcsum;	/* # of times hardware assisted with checksum */
619 	uint64_t tso_wrs;	/* # of TSO work requests */
620 	uint64_t vlan_insertion;/* # of times VLAN tag was inserted */
621 	uint64_t imm_wrs;	/* # of work requests with immediate data */
622 	uint64_t sgl_wrs;	/* # of work requests with direct SGL */
623 	uint64_t txpkt_wrs;	/* # of txpkt work requests (not coalesced) */
624 	uint64_t txpkts0_wrs;	/* # of type0 coalesced tx work requests */
625 	uint64_t txpkts1_wrs;	/* # of type1 coalesced tx work requests */
626 	uint64_t txpkts0_pkts;	/* # of frames in type0 coalesced tx WRs */
627 	uint64_t txpkts1_pkts;	/* # of frames in type1 coalesced tx WRs */
628 	uint64_t txpkts_flush;	/* # of times txp had to be sent by tx_update */
629 	uint64_t raw_wrs;	/* # of raw work requests (alloc_wr_mbuf) */
630 	uint64_t vxlan_tso_wrs;	/* # of VXLAN TSO work requests */
631 	uint64_t vxlan_txcsum;
632 
633 	uint64_t kern_tls_records;
634 	uint64_t kern_tls_short;
635 	uint64_t kern_tls_partial;
636 	uint64_t kern_tls_full;
637 	uint64_t kern_tls_octets;
638 	uint64_t kern_tls_waste;
639 	uint64_t kern_tls_options;
640 	uint64_t kern_tls_header;
641 	uint64_t kern_tls_fin;
642 	uint64_t kern_tls_fin_short;
643 	uint64_t kern_tls_cbc;
644 	uint64_t kern_tls_gcm;
645 
646 	/* stats for not-that-common events */
647 
648 	/* Optional scratch space for constructing work requests. */
649 	uint8_t ss[SGE_MAX_WR_LEN] __aligned(16);
650 } __aligned(CACHE_LINE_SIZE);
651 
652 /* rxq: SGE ingress queue + SGE free list + miscellaneous items */
653 struct sge_rxq {
654 	struct sge_iq iq;	/* MUST be first */
655 	struct sge_fl fl;	/* MUST follow iq */
656 
657 	if_t ifp;		/* the interface this rxq belongs to */
658 	struct lro_ctrl lro;	/* LRO state */
659 
660 	/* stats for common events first */
661 
662 	uint64_t rxcsum;	/* # of times hardware assisted with checksum */
663 	uint64_t vlan_extraction;/* # of times VLAN tag was extracted */
664 	uint64_t vxlan_rxcsum;
665 
666 	/* stats for not-that-common events */
667 
668 } __aligned(CACHE_LINE_SIZE);
669 
670 static inline struct sge_rxq *
671 iq_to_rxq(struct sge_iq *iq)
672 {
673 
674 	return (__containerof(iq, struct sge_rxq, iq));
675 }
676 
677 /* ofld_rxq: SGE ingress queue + SGE free list + miscellaneous items */
678 struct sge_ofld_rxq {
679 	struct sge_iq iq;	/* MUST be first */
680 	struct sge_fl fl;	/* MUST follow iq */
681 	counter_u64_t rx_iscsi_ddp_setup_ok;
682 	counter_u64_t rx_iscsi_ddp_setup_error;
683 	uint64_t rx_iscsi_ddp_pdus;
684 	uint64_t rx_iscsi_ddp_octets;
685 	uint64_t rx_iscsi_fl_pdus;
686 	uint64_t rx_iscsi_fl_octets;
687 	uint64_t rx_iscsi_padding_errors;
688 	uint64_t rx_iscsi_header_digest_errors;
689 	uint64_t rx_iscsi_data_digest_errors;
690 	u_long	rx_toe_tls_records;
691 	u_long	rx_toe_tls_octets;
692 } __aligned(CACHE_LINE_SIZE);
693 
694 static inline struct sge_ofld_rxq *
695 iq_to_ofld_rxq(struct sge_iq *iq)
696 {
697 
698 	return (__containerof(iq, struct sge_ofld_rxq, iq));
699 }
700 
701 struct wrqe {
702 	STAILQ_ENTRY(wrqe) link;
703 	struct sge_wrq *wrq;
704 	int wr_len;
705 	char wr[] __aligned(16);
706 };
707 
708 struct wrq_cookie {
709 	TAILQ_ENTRY(wrq_cookie) link;
710 	int ndesc;
711 	int pidx;
712 };
713 
714 /*
715  * wrq: SGE egress queue that is given prebuilt work requests.  Control queues
716  * are of this type.
717  */
718 struct sge_wrq {
719 	struct sge_eq eq;	/* MUST be first */
720 
721 	struct adapter *adapter;
722 	struct task wrq_tx_task;
723 
724 	/* Tx desc reserved but WR not "committed" yet. */
725 	TAILQ_HEAD(wrq_incomplete_wrs , wrq_cookie) incomplete_wrs;
726 
727 	/* List of WRs ready to go out as soon as descriptors are available. */
728 	STAILQ_HEAD(, wrqe) wr_list;
729 	u_int nwr_pending;
730 	u_int ndesc_needed;
731 
732 	/* stats for common events first */
733 
734 	uint64_t tx_wrs_direct;	/* # of WRs written directly to desc ring. */
735 	uint64_t tx_wrs_ss;	/* # of WRs copied from scratch space. */
736 	uint64_t tx_wrs_copied;	/* # of WRs queued and copied to desc ring. */
737 
738 	/* stats for not-that-common events */
739 
740 	/*
741 	 * Scratch space for work requests that wrap around after reaching the
742 	 * status page, and some information about the last WR that used it.
743 	 */
744 	uint16_t ss_pidx;
745 	uint16_t ss_len;
746 	uint8_t ss[SGE_MAX_WR_LEN];
747 
748 } __aligned(CACHE_LINE_SIZE);
749 
750 /* ofld_txq: SGE egress queue + miscellaneous items */
751 struct sge_ofld_txq {
752 	struct sge_wrq wrq;
753 	counter_u64_t tx_iscsi_pdus;
754 	counter_u64_t tx_iscsi_octets;
755 	counter_u64_t tx_iscsi_iso_wrs;
756 	counter_u64_t tx_toe_tls_records;
757 	counter_u64_t tx_toe_tls_octets;
758 } __aligned(CACHE_LINE_SIZE);
759 
760 #define INVALID_NM_RXQ_CNTXT_ID ((uint16_t)(-1))
761 struct sge_nm_rxq {
762 	/* Items used by the driver rx ithread are in this cacheline. */
763 	volatile int nm_state __aligned(CACHE_LINE_SIZE);	/* NM_OFF, NM_ON, or NM_BUSY */
764 	u_int nid;		/* netmap ring # for this queue */
765 	struct vi_info *vi;
766 
767 	struct iq_desc *iq_desc;
768 	uint16_t iq_abs_id;
769 	uint16_t iq_cntxt_id;
770 	uint16_t iq_cidx;
771 	uint16_t iq_sidx;
772 	uint8_t iq_gen;
773 	uint32_t fl_sidx;
774 
775 	/* Items used by netmap rxsync are in this cacheline. */
776 	__be64  *fl_desc __aligned(CACHE_LINE_SIZE);
777 	uint16_t fl_cntxt_id;
778 	uint32_t fl_pidx;
779 	uint32_t fl_sidx2;	/* copy of fl_sidx */
780 	uint32_t fl_db_val;
781 	u_int fl_db_saved;
782 	u_int fl_db_threshold;	/* in descriptors */
783 	u_int fl_hwidx:4;
784 
785 	/*
786 	 * fl_cidx is used by both the ithread and rxsync, the rest are not used
787 	 * in the rx fast path.
788 	 */
789 	uint32_t fl_cidx __aligned(CACHE_LINE_SIZE);
790 
791 	bus_dma_tag_t iq_desc_tag;
792 	bus_dmamap_t iq_desc_map;
793 	bus_addr_t iq_ba;
794 	int intr_idx;
795 
796 	bus_dma_tag_t fl_desc_tag;
797 	bus_dmamap_t fl_desc_map;
798 	bus_addr_t fl_ba;
799 };
800 
801 #define INVALID_NM_TXQ_CNTXT_ID ((u_int)(-1))
802 struct sge_nm_txq {
803 	struct tx_desc *desc;
804 	uint16_t cidx;
805 	uint16_t pidx;
806 	uint16_t sidx;
807 	uint16_t equiqidx;	/* EQUIQ last requested at this pidx */
808 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
809 	uint16_t dbidx;		/* pidx of the most recent doorbell */
810 	uint8_t doorbells;
811 	volatile uint32_t *udb;
812 	u_int udb_qid;
813 	u_int cntxt_id;
814 	__be32 cpl_ctrl0;	/* for convenience */
815 	__be32 op_pkd;		/* ditto */
816 	u_int nid;		/* netmap ring # for this queue */
817 
818 	/* infrequently used items after this */
819 
820 	bus_dma_tag_t desc_tag;
821 	bus_dmamap_t desc_map;
822 	bus_addr_t ba;
823 	int iqidx;
824 } __aligned(CACHE_LINE_SIZE);
825 
826 struct sge {
827 	int nrxq;	/* total # of Ethernet rx queues */
828 	int ntxq;	/* total # of Ethernet tx queues */
829 	int nofldrxq;	/* total # of TOE rx queues */
830 	int nofldtxq;	/* total # of TOE tx queues */
831 	int nnmrxq;	/* total # of netmap rx queues */
832 	int nnmtxq;	/* total # of netmap tx queues */
833 	int niq;	/* total # of ingress queues */
834 	int neq;	/* total # of egress queues */
835 
836 	struct sge_iq fwq;	/* Firmware event queue */
837 	struct sge_wrq *ctrlq;	/* Control queues */
838 	struct sge_txq *txq;	/* NIC tx queues */
839 	struct sge_rxq *rxq;	/* NIC rx queues */
840 	struct sge_ofld_txq *ofld_txq;	/* TOE tx queues */
841 	struct sge_ofld_rxq *ofld_rxq;	/* TOE rx queues */
842 	struct sge_nm_txq *nm_txq;	/* netmap tx queues */
843 	struct sge_nm_rxq *nm_rxq;	/* netmap rx queues */
844 
845 	uint16_t iq_start;	/* first cntxt_id */
846 	uint16_t iq_base;	/* first abs_id */
847 	int eq_start;		/* first cntxt_id */
848 	int eq_base;		/* first abs_id */
849 	int iqmap_sz;
850 	int eqmap_sz;
851 	struct sge_iq **iqmap;	/* iq->cntxt_id to iq mapping */
852 	struct sge_eq **eqmap;	/* eq->cntxt_id to eq mapping */
853 
854 	int8_t safe_zidx;
855 	struct rx_buf_info rx_buf_info[SW_ZONE_SIZES];
856 };
857 
858 struct devnames {
859 	const char *nexus_name;
860 	const char *ifnet_name;
861 	const char *vi_ifnet_name;
862 	const char *pf03_drv_name;
863 	const char *vf_nexus_name;
864 	const char *vf_ifnet_name;
865 };
866 
867 struct clip_entry;
868 
869 #define CNT_CAL_INFO 3
870 struct clock_sync {
871 	uint64_t hw_cur;
872 	uint64_t hw_prev;
873 	sbintime_t sbt_cur;
874 	sbintime_t sbt_prev;
875 	seqc_t gen;
876 };
877 
878 struct adapter {
879 	SLIST_ENTRY(adapter) link;
880 	device_t dev;
881 	struct cdev *cdev;
882 	const struct devnames *names;
883 
884 	/* PCIe register resources */
885 	int regs_rid;
886 	struct resource *regs_res;
887 	int msix_rid;
888 	struct resource *msix_res;
889 	bus_space_handle_t bh;
890 	bus_space_tag_t bt;
891 	bus_size_t mmio_len;
892 	int udbs_rid;
893 	struct resource *udbs_res;
894 	volatile uint8_t *udbs_base;
895 
896 	unsigned int pf;
897 	unsigned int mbox;
898 	unsigned int vpd_busy;
899 	unsigned int vpd_flag;
900 
901 	/* Interrupt information */
902 	int intr_type;
903 	int intr_count;
904 	struct irq {
905 		struct resource *res;
906 		int rid;
907 		void *tag;
908 		struct sge_rxq *rxq;
909 		struct sge_nm_rxq *nm_rxq;
910 	} __aligned(CACHE_LINE_SIZE) *irq;
911 	int sge_gts_reg;
912 	int sge_kdoorbell_reg;
913 
914 	bus_dma_tag_t dmat;	/* Parent DMA tag */
915 
916 	struct sge sge;
917 	int lro_timeout;
918 	int sc_do_rxcopy;
919 
920 	int vxlan_port;
921 	u_int vxlan_refcount;
922 	int rawf_base;
923 	int nrawf;
924 
925 	struct taskqueue *tq[MAX_NCHAN];	/* General purpose taskqueues */
926 	struct port_info *port[MAX_NPORTS];
927 	uint8_t chan_map[MAX_NCHAN];		/* channel -> port */
928 
929 	CXGBE_LIST_HEAD(, clip_entry) *clip_table;
930 	TAILQ_HEAD(, clip_entry) clip_pending;	/* these need hw update. */
931 	u_long clip_mask;
932 	int clip_gen;
933 	struct timeout_task clip_task;
934 
935 	void *tom_softc;	/* (struct tom_data *) */
936 	struct tom_tunables tt;
937 	struct t4_offload_policy *policy;
938 	struct rwlock policy_lock;
939 
940 	void *iwarp_softc;	/* (struct c4iw_dev *) */
941 	struct iw_tunables iwt;
942 	void *iscsi_ulp_softc;	/* (struct cxgbei_data *) */
943 	struct l2t_data *l2t;	/* L2 table */
944 	struct smt_data *smt;	/* Source MAC Table */
945 	struct tid_info tids;
946 	vmem_t *key_map;
947 	struct tls_tunables tlst;
948 
949 	uint8_t doorbells;
950 	int offload_map;	/* port_id's with IFCAP_TOE enabled */
951 	int bt_map;		/* tx_chan's with BASE-T */
952 	int active_ulds;	/* ULDs activated on this adapter */
953 	int flags;
954 	int debug_flags;
955 	int error_flags;	/* Used by error handler and live reset. */
956 
957 	char ifp_lockname[16];
958 	struct mtx ifp_lock;
959 	if_t ifp;		/* tracer ifp */
960 	struct ifmedia media;
961 	int traceq;		/* iq used by all tracers, -1 if none */
962 	int tracer_valid;	/* bitmap of valid tracers */
963 	int tracer_enabled;	/* bitmap of enabled tracers */
964 
965 	char fw_version[16];
966 	char tp_version[16];
967 	char er_version[16];
968 	char bs_version[16];
969 	char cfg_file[32];
970 	u_int cfcsum;
971 	struct adapter_params params;
972 	const struct chip_params *chip_params;
973 	struct t4_virt_res vres;
974 
975 	uint16_t nbmcaps;
976 	uint16_t linkcaps;
977 	uint16_t switchcaps;
978 	uint16_t niccaps;
979 	uint16_t toecaps;
980 	uint16_t rdmacaps;
981 	uint16_t cryptocaps;
982 	uint16_t iscsicaps;
983 	uint16_t fcoecaps;
984 
985 	struct sysctl_ctx_list ctx;
986 	struct sysctl_oid *ctrlq_oid;
987 	struct sysctl_oid *fwq_oid;
988 
989 	struct mtx sc_lock;
990 	char lockname[16];
991 
992 	/* Starving free lists */
993 	struct mtx sfl_lock;	/* same cache-line as sc_lock? but that's ok */
994 	TAILQ_HEAD(, sge_fl) sfl;
995 	struct callout sfl_callout;
996 	struct callout cal_callout;
997 	struct clock_sync cal_info[CNT_CAL_INFO];
998 	int cal_current;
999 	int cal_count;
1000 	uint32_t cal_gen;
1001 
1002 	/*
1003 	 * Driver code that can run when the adapter is suspended must use this
1004 	 * lock or a synchronized_op and check for HW_OFF_LIMITS before
1005 	 * accessing hardware.
1006 	 *
1007 	 * XXX: could be changed to rwlock.  wlock in suspend/resume and for
1008 	 * indirect register access, rlock everywhere else.
1009 	 */
1010 	struct mtx reg_lock;
1011 
1012 	struct memwin memwin[NUM_MEMWIN];	/* memory windows */
1013 
1014 	struct mtx tc_lock;
1015 	struct task tc_task;
1016 
1017 	struct task fatal_error_task;
1018 	struct task reset_task;
1019 	const void *reset_thread;
1020 	int num_resets;
1021 	int incarnation;
1022 
1023 	const char *last_op;
1024 	const void *last_op_thr;
1025 	int last_op_flags;
1026 
1027 	int swintr;
1028 	int sensor_resets;
1029 
1030 	struct callout ktls_tick;
1031 };
1032 
1033 #define ADAPTER_LOCK(sc)		mtx_lock(&(sc)->sc_lock)
1034 #define ADAPTER_UNLOCK(sc)		mtx_unlock(&(sc)->sc_lock)
1035 #define ADAPTER_LOCK_ASSERT_OWNED(sc)	mtx_assert(&(sc)->sc_lock, MA_OWNED)
1036 #define ADAPTER_LOCK_ASSERT_NOTOWNED(sc) mtx_assert(&(sc)->sc_lock, MA_NOTOWNED)
1037 
1038 #define ASSERT_SYNCHRONIZED_OP(sc)	\
1039     KASSERT(IS_BUSY(sc) && \
1040 	(mtx_owned(&(sc)->sc_lock) || sc->last_op_thr == curthread), \
1041 	("%s: operation not synchronized.", __func__))
1042 
1043 #define PORT_LOCK(pi)			mtx_lock(&(pi)->pi_lock)
1044 #define PORT_UNLOCK(pi)			mtx_unlock(&(pi)->pi_lock)
1045 #define PORT_LOCK_ASSERT_OWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_OWNED)
1046 #define PORT_LOCK_ASSERT_NOTOWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_NOTOWNED)
1047 
1048 #define FL_LOCK(fl)			mtx_lock(&(fl)->fl_lock)
1049 #define FL_TRYLOCK(fl)			mtx_trylock(&(fl)->fl_lock)
1050 #define FL_UNLOCK(fl)			mtx_unlock(&(fl)->fl_lock)
1051 #define FL_LOCK_ASSERT_OWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_OWNED)
1052 #define FL_LOCK_ASSERT_NOTOWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_NOTOWNED)
1053 
1054 #define RXQ_FL_LOCK(rxq)		FL_LOCK(&(rxq)->fl)
1055 #define RXQ_FL_UNLOCK(rxq)		FL_UNLOCK(&(rxq)->fl)
1056 #define RXQ_FL_LOCK_ASSERT_OWNED(rxq)	FL_LOCK_ASSERT_OWNED(&(rxq)->fl)
1057 #define RXQ_FL_LOCK_ASSERT_NOTOWNED(rxq) FL_LOCK_ASSERT_NOTOWNED(&(rxq)->fl)
1058 
1059 #define EQ_LOCK(eq)			mtx_lock(&(eq)->eq_lock)
1060 #define EQ_TRYLOCK(eq)			mtx_trylock(&(eq)->eq_lock)
1061 #define EQ_UNLOCK(eq)			mtx_unlock(&(eq)->eq_lock)
1062 #define EQ_LOCK_ASSERT_OWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_OWNED)
1063 #define EQ_LOCK_ASSERT_NOTOWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_NOTOWNED)
1064 
1065 #define TXQ_LOCK(txq)			EQ_LOCK(&(txq)->eq)
1066 #define TXQ_TRYLOCK(txq)		EQ_TRYLOCK(&(txq)->eq)
1067 #define TXQ_UNLOCK(txq)			EQ_UNLOCK(&(txq)->eq)
1068 #define TXQ_LOCK_ASSERT_OWNED(txq)	EQ_LOCK_ASSERT_OWNED(&(txq)->eq)
1069 #define TXQ_LOCK_ASSERT_NOTOWNED(txq)	EQ_LOCK_ASSERT_NOTOWNED(&(txq)->eq)
1070 
1071 #define for_each_txq(vi, iter, q) \
1072 	for (q = &vi->adapter->sge.txq[vi->first_txq], iter = 0; \
1073 	    iter < vi->ntxq; ++iter, ++q)
1074 #define for_each_rxq(vi, iter, q) \
1075 	for (q = &vi->adapter->sge.rxq[vi->first_rxq], iter = 0; \
1076 	    iter < vi->nrxq; ++iter, ++q)
1077 #define for_each_ofld_txq(vi, iter, q) \
1078 	for (q = &vi->adapter->sge.ofld_txq[vi->first_ofld_txq], iter = 0; \
1079 	    iter < vi->nofldtxq; ++iter, ++q)
1080 #define for_each_ofld_rxq(vi, iter, q) \
1081 	for (q = &vi->adapter->sge.ofld_rxq[vi->first_ofld_rxq], iter = 0; \
1082 	    iter < vi->nofldrxq; ++iter, ++q)
1083 #define for_each_nm_txq(vi, iter, q) \
1084 	for (q = &vi->adapter->sge.nm_txq[vi->first_nm_txq], iter = 0; \
1085 	    iter < vi->nnmtxq; ++iter, ++q)
1086 #define for_each_nm_rxq(vi, iter, q) \
1087 	for (q = &vi->adapter->sge.nm_rxq[vi->first_nm_rxq], iter = 0; \
1088 	    iter < vi->nnmrxq; ++iter, ++q)
1089 #define for_each_vi(_pi, _iter, _vi) \
1090 	for ((_vi) = (_pi)->vi, (_iter) = 0; (_iter) < (_pi)->nvi; \
1091 	     ++(_iter), ++(_vi))
1092 
1093 #define IDXINCR(idx, incr, wrap) do { \
1094 	idx = wrap - idx > incr ? idx + incr : incr - (wrap - idx); \
1095 } while (0)
1096 #define IDXDIFF(head, tail, wrap) \
1097 	((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head))
1098 
1099 /* One for errors, one for firmware events */
1100 #define T4_EXTRA_INTR 2
1101 
1102 /* One for firmware events */
1103 #define T4VF_EXTRA_INTR 1
1104 
1105 static inline int
1106 forwarding_intr_to_fwq(struct adapter *sc)
1107 {
1108 
1109 	return (sc->intr_count == 1);
1110 }
1111 
1112 /* Works reliably inside a sync_op or with reg_lock held. */
1113 static inline bool
1114 hw_off_limits(struct adapter *sc)
1115 {
1116 	int off_limits = atomic_load_int(&sc->error_flags) & HW_OFF_LIMITS;
1117 
1118 	return (__predict_false(off_limits != 0));
1119 }
1120 
1121 static inline int
1122 mbuf_nsegs(struct mbuf *m)
1123 {
1124 	M_ASSERTPKTHDR(m);
1125 	KASSERT(m->m_pkthdr.inner_l5hlen > 0,
1126 	    ("%s: mbuf %p missing information on # of segments.", __func__, m));
1127 
1128 	return (m->m_pkthdr.inner_l5hlen);
1129 }
1130 
1131 static inline void
1132 set_mbuf_nsegs(struct mbuf *m, uint8_t nsegs)
1133 {
1134 	M_ASSERTPKTHDR(m);
1135 	m->m_pkthdr.inner_l5hlen = nsegs;
1136 }
1137 
1138 /* Internal mbuf flags stored in PH_loc.eight[1]. */
1139 #define	MC_NOMAP		0x01
1140 #define	MC_RAW_WR		0x02
1141 #define	MC_TLS			0x04
1142 
1143 static inline int
1144 mbuf_cflags(struct mbuf *m)
1145 {
1146 	M_ASSERTPKTHDR(m);
1147 	return (m->m_pkthdr.PH_loc.eight[4]);
1148 }
1149 
1150 static inline void
1151 set_mbuf_cflags(struct mbuf *m, uint8_t flags)
1152 {
1153 	M_ASSERTPKTHDR(m);
1154 	m->m_pkthdr.PH_loc.eight[4] = flags;
1155 }
1156 
1157 static inline int
1158 mbuf_len16(struct mbuf *m)
1159 {
1160 	int n;
1161 
1162 	M_ASSERTPKTHDR(m);
1163 	n = m->m_pkthdr.PH_loc.eight[0];
1164 	if (!(mbuf_cflags(m) & MC_TLS))
1165 		MPASS(n > 0 && n <= SGE_MAX_WR_LEN / 16);
1166 
1167 	return (n);
1168 }
1169 
1170 static inline void
1171 set_mbuf_len16(struct mbuf *m, uint8_t len16)
1172 {
1173 	M_ASSERTPKTHDR(m);
1174 	if (!(mbuf_cflags(m) & MC_TLS))
1175 		MPASS(len16 > 0 && len16 <= SGE_MAX_WR_LEN / 16);
1176 	m->m_pkthdr.PH_loc.eight[0] = len16;
1177 }
1178 
1179 static inline uint32_t
1180 t4_read_reg(struct adapter *sc, uint32_t reg)
1181 {
1182 	if (hw_off_limits(sc))
1183 		MPASS(curthread == sc->reset_thread);
1184 	return bus_space_read_4(sc->bt, sc->bh, reg);
1185 }
1186 
1187 static inline void
1188 t4_write_reg(struct adapter *sc, uint32_t reg, uint32_t val)
1189 {
1190 	if (hw_off_limits(sc))
1191 		MPASS(curthread == sc->reset_thread);
1192 	bus_space_write_4(sc->bt, sc->bh, reg, val);
1193 }
1194 
1195 static inline uint64_t
1196 t4_read_reg64(struct adapter *sc, uint32_t reg)
1197 {
1198 	if (hw_off_limits(sc))
1199 		MPASS(curthread == sc->reset_thread);
1200 #ifdef __LP64__
1201 	return bus_space_read_8(sc->bt, sc->bh, reg);
1202 #else
1203 	return (uint64_t)bus_space_read_4(sc->bt, sc->bh, reg) +
1204 	    ((uint64_t)bus_space_read_4(sc->bt, sc->bh, reg + 4) << 32);
1205 
1206 #endif
1207 }
1208 
1209 static inline void
1210 t4_write_reg64(struct adapter *sc, uint32_t reg, uint64_t val)
1211 {
1212 	if (hw_off_limits(sc))
1213 		MPASS(curthread == sc->reset_thread);
1214 #ifdef __LP64__
1215 	bus_space_write_8(sc->bt, sc->bh, reg, val);
1216 #else
1217 	bus_space_write_4(sc->bt, sc->bh, reg, val);
1218 	bus_space_write_4(sc->bt, sc->bh, reg + 4, val>> 32);
1219 #endif
1220 }
1221 
1222 static inline void
1223 t4_os_pci_read_cfg1(struct adapter *sc, int reg, uint8_t *val)
1224 {
1225 	if (hw_off_limits(sc))
1226 		MPASS(curthread == sc->reset_thread);
1227 	*val = pci_read_config(sc->dev, reg, 1);
1228 }
1229 
1230 static inline void
1231 t4_os_pci_write_cfg1(struct adapter *sc, int reg, uint8_t val)
1232 {
1233 	if (hw_off_limits(sc))
1234 		MPASS(curthread == sc->reset_thread);
1235 	pci_write_config(sc->dev, reg, val, 1);
1236 }
1237 
1238 static inline void
1239 t4_os_pci_read_cfg2(struct adapter *sc, int reg, uint16_t *val)
1240 {
1241 
1242 	if (hw_off_limits(sc))
1243 		MPASS(curthread == sc->reset_thread);
1244 	*val = pci_read_config(sc->dev, reg, 2);
1245 }
1246 
1247 static inline void
1248 t4_os_pci_write_cfg2(struct adapter *sc, int reg, uint16_t val)
1249 {
1250 	if (hw_off_limits(sc))
1251 		MPASS(curthread == sc->reset_thread);
1252 	pci_write_config(sc->dev, reg, val, 2);
1253 }
1254 
1255 static inline void
1256 t4_os_pci_read_cfg4(struct adapter *sc, int reg, uint32_t *val)
1257 {
1258 	if (hw_off_limits(sc))
1259 		MPASS(curthread == sc->reset_thread);
1260 	*val = pci_read_config(sc->dev, reg, 4);
1261 }
1262 
1263 static inline void
1264 t4_os_pci_write_cfg4(struct adapter *sc, int reg, uint32_t val)
1265 {
1266 	if (hw_off_limits(sc))
1267 		MPASS(curthread == sc->reset_thread);
1268 	pci_write_config(sc->dev, reg, val, 4);
1269 }
1270 
1271 static inline struct port_info *
1272 adap2pinfo(struct adapter *sc, int idx)
1273 {
1274 
1275 	return (sc->port[idx]);
1276 }
1277 
1278 static inline void
1279 t4_os_set_hw_addr(struct port_info *pi, uint8_t hw_addr[])
1280 {
1281 
1282 	bcopy(hw_addr, pi->vi[0].hw_addr, ETHER_ADDR_LEN);
1283 }
1284 
1285 static inline int
1286 tx_resume_threshold(struct sge_eq *eq)
1287 {
1288 
1289 	/* not quite the same as qsize / 4, but this will do. */
1290 	return (eq->sidx / 4);
1291 }
1292 
1293 static inline int
1294 t4_use_ldst(struct adapter *sc)
1295 {
1296 
1297 #ifdef notyet
1298 	return (sc->flags & FW_OK || !sc->use_bd);
1299 #else
1300 	return (0);
1301 #endif
1302 }
1303 
1304 static inline void
1305 CH_DUMP_MBOX(struct adapter *sc, int mbox, const int reg,
1306     const char *msg, const __be64 *const p, const bool err)
1307 {
1308 
1309 	if (!(sc->debug_flags & DF_DUMP_MBOX) && !err)
1310 		return;
1311 	if (p != NULL) {
1312 		log(err ? LOG_ERR : LOG_DEBUG,
1313 		    "%s: mbox %u %s %016llx %016llx %016llx %016llx "
1314 		    "%016llx %016llx %016llx %016llx\n",
1315 		    device_get_nameunit(sc->dev), mbox, msg,
1316 		    (long long)be64_to_cpu(p[0]), (long long)be64_to_cpu(p[1]),
1317 		    (long long)be64_to_cpu(p[2]), (long long)be64_to_cpu(p[3]),
1318 		    (long long)be64_to_cpu(p[4]), (long long)be64_to_cpu(p[5]),
1319 		    (long long)be64_to_cpu(p[6]), (long long)be64_to_cpu(p[7]));
1320 	} else {
1321 		log(err ? LOG_ERR : LOG_DEBUG,
1322 		    "%s: mbox %u %s %016llx %016llx %016llx %016llx "
1323 		    "%016llx %016llx %016llx %016llx\n",
1324 		    device_get_nameunit(sc->dev), mbox, msg,
1325 		    (long long)t4_read_reg64(sc, reg),
1326 		    (long long)t4_read_reg64(sc, reg + 8),
1327 		    (long long)t4_read_reg64(sc, reg + 16),
1328 		    (long long)t4_read_reg64(sc, reg + 24),
1329 		    (long long)t4_read_reg64(sc, reg + 32),
1330 		    (long long)t4_read_reg64(sc, reg + 40),
1331 		    (long long)t4_read_reg64(sc, reg + 48),
1332 		    (long long)t4_read_reg64(sc, reg + 56));
1333 	}
1334 }
1335 
1336 /* t4_main.c */
1337 extern int t4_ntxq;
1338 extern int t4_nrxq;
1339 extern int t4_intr_types;
1340 extern int t4_tmr_idx;
1341 extern int t4_pktc_idx;
1342 extern unsigned int t4_qsize_rxq;
1343 extern unsigned int t4_qsize_txq;
1344 extern device_method_t cxgbe_methods[];
1345 
1346 int t4_os_find_pci_capability(struct adapter *, int);
1347 int t4_os_pci_save_state(struct adapter *);
1348 int t4_os_pci_restore_state(struct adapter *);
1349 void t4_os_portmod_changed(struct port_info *);
1350 void t4_os_link_changed(struct port_info *);
1351 void t4_iterate(void (*)(struct adapter *, void *), void *);
1352 void t4_init_devnames(struct adapter *);
1353 void t4_add_adapter(struct adapter *);
1354 int t4_detach_common(device_t);
1355 int t4_map_bars_0_and_4(struct adapter *);
1356 int t4_map_bar_2(struct adapter *);
1357 int t4_setup_intr_handlers(struct adapter *);
1358 void t4_sysctls(struct adapter *);
1359 int begin_synchronized_op(struct adapter *, struct vi_info *, int, char *);
1360 void doom_vi(struct adapter *, struct vi_info *);
1361 void end_synchronized_op(struct adapter *, int);
1362 int update_mac_settings(if_t, int);
1363 int adapter_init(struct adapter *);
1364 int vi_init(struct vi_info *);
1365 void vi_sysctls(struct vi_info *);
1366 int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
1367 int alloc_atid(struct adapter *, void *);
1368 void *lookup_atid(struct adapter *, int);
1369 void free_atid(struct adapter *, int);
1370 void release_tid(struct adapter *, int, struct sge_wrq *);
1371 int cxgbe_media_change(if_t);
1372 void cxgbe_media_status(if_t, struct ifmediareq *);
1373 void t4_os_cim_err(struct adapter *);
1374 
1375 #ifdef KERN_TLS
1376 /* t6_kern_tls.c */
1377 int t6_tls_tag_alloc(if_t, union if_snd_tag_alloc_params *,
1378     struct m_snd_tag **);
1379 void t6_ktls_modload(void);
1380 void t6_ktls_modunload(void);
1381 int t6_ktls_try(if_t, struct socket *, struct ktls_session *);
1382 int t6_ktls_parse_pkt(struct mbuf *);
1383 int t6_ktls_write_wr(struct sge_txq *, void *, struct mbuf *, u_int);
1384 #endif
1385 
1386 /* t4_keyctx.c */
1387 struct auth_hash;
1388 union authctx;
1389 #ifdef KERN_TLS
1390 struct ktls_session;
1391 struct tls_key_req;
1392 struct tls_keyctx;
1393 #endif
1394 
1395 void t4_aes_getdeckey(void *, const void *, unsigned int);
1396 void t4_copy_partial_hash(int, union authctx *, void *);
1397 void t4_init_gmac_hash(const char *, int, char *);
1398 void t4_init_hmac_digest(const struct auth_hash *, u_int, const char *, int,
1399     char *);
1400 #ifdef KERN_TLS
1401 u_int t4_tls_key_info_size(const struct ktls_session *);
1402 int t4_tls_proto_ver(const struct ktls_session *);
1403 int t4_tls_cipher_mode(const struct ktls_session *);
1404 int t4_tls_auth_mode(const struct ktls_session *);
1405 int t4_tls_hmac_ctrl(const struct ktls_session *);
1406 void t4_tls_key_ctx(const struct ktls_session *, int, struct tls_keyctx *);
1407 int t4_alloc_tls_keyid(struct adapter *);
1408 void t4_free_tls_keyid(struct adapter *, int);
1409 void t4_write_tlskey_wr(const struct ktls_session *, int, int, int, int,
1410     struct tls_key_req *);
1411 #endif
1412 
1413 #ifdef DEV_NETMAP
1414 /* t4_netmap.c */
1415 struct sge_nm_rxq;
1416 void cxgbe_nm_attach(struct vi_info *);
1417 void cxgbe_nm_detach(struct vi_info *);
1418 void service_nm_rxq(struct sge_nm_rxq *);
1419 int alloc_nm_rxq(struct vi_info *, struct sge_nm_rxq *, int, int);
1420 int free_nm_rxq(struct vi_info *, struct sge_nm_rxq *);
1421 int alloc_nm_txq(struct vi_info *, struct sge_nm_txq *, int, int);
1422 int free_nm_txq(struct vi_info *, struct sge_nm_txq *);
1423 #endif
1424 
1425 /* t4_sge.c */
1426 void t4_sge_modload(void);
1427 void t4_sge_modunload(void);
1428 uint64_t t4_sge_extfree_refs(void);
1429 void t4_tweak_chip_settings(struct adapter *);
1430 int t4_verify_chip_settings(struct adapter *);
1431 void t4_init_rx_buf_info(struct adapter *);
1432 int t4_create_dma_tag(struct adapter *);
1433 void t4_sge_sysctls(struct adapter *, struct sysctl_ctx_list *,
1434     struct sysctl_oid_list *);
1435 int t4_destroy_dma_tag(struct adapter *);
1436 int alloc_ring(struct adapter *, size_t, bus_dma_tag_t *, bus_dmamap_t *,
1437     bus_addr_t *, void **);
1438 int free_ring(struct adapter *, bus_dma_tag_t, bus_dmamap_t, bus_addr_t,
1439     void *);
1440 void free_fl_buffers(struct adapter *, struct sge_fl *);
1441 int t4_setup_adapter_queues(struct adapter *);
1442 int t4_teardown_adapter_queues(struct adapter *);
1443 int t4_setup_vi_queues(struct vi_info *);
1444 int t4_teardown_vi_queues(struct vi_info *);
1445 void t4_intr_all(void *);
1446 void t4_intr(void *);
1447 #ifdef DEV_NETMAP
1448 void t4_nm_intr(void *);
1449 void t4_vi_intr(void *);
1450 #endif
1451 void t4_intr_err(void *);
1452 void t4_intr_evt(void *);
1453 void t4_wrq_tx_locked(struct adapter *, struct sge_wrq *, struct wrqe *);
1454 void t4_update_fl_bufsize(if_t);
1455 struct mbuf *alloc_wr_mbuf(int, int);
1456 int parse_pkt(struct mbuf **, bool);
1457 void *start_wrq_wr(struct sge_wrq *, int, struct wrq_cookie *);
1458 void commit_wrq_wr(struct sge_wrq *, void *, struct wrq_cookie *);
1459 int t4_sge_set_conm_context(struct adapter *, int, int, int);
1460 void t4_register_an_handler(an_handler_t);
1461 void t4_register_fw_msg_handler(int, fw_msg_handler_t);
1462 void t4_register_cpl_handler(int, cpl_handler_t);
1463 void t4_register_shared_cpl_handler(int, cpl_handler_t, int);
1464 #ifdef RATELIMIT
1465 void send_etid_flush_wr(struct cxgbe_rate_tag *);
1466 #endif
1467 
1468 /* t4_tracer.c */
1469 struct t4_tracer;
1470 void t4_tracer_modload(void);
1471 void t4_tracer_modunload(void);
1472 void t4_tracer_port_detach(struct adapter *);
1473 int t4_get_tracer(struct adapter *, struct t4_tracer *);
1474 int t4_set_tracer(struct adapter *, struct t4_tracer *);
1475 int t4_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1476 int t5_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1477 
1478 /* t4_sched.c */
1479 int t4_set_sched_class(struct adapter *, struct t4_sched_params *);
1480 int t4_set_sched_queue(struct adapter *, struct t4_sched_queue *);
1481 int t4_init_tx_sched(struct adapter *);
1482 int t4_free_tx_sched(struct adapter *);
1483 void t4_update_tx_sched(struct adapter *);
1484 int t4_reserve_cl_rl_kbps(struct adapter *, int, u_int, int *);
1485 void t4_release_cl_rl(struct adapter *, int, int);
1486 int sysctl_tc(SYSCTL_HANDLER_ARGS);
1487 int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
1488 #ifdef RATELIMIT
1489 void t4_init_etid_table(struct adapter *);
1490 void t4_free_etid_table(struct adapter *);
1491 struct cxgbe_rate_tag *lookup_etid(struct adapter *, int);
1492 int cxgbe_rate_tag_alloc(if_t, union if_snd_tag_alloc_params *,
1493     struct m_snd_tag **);
1494 void cxgbe_rate_tag_free_locked(struct cxgbe_rate_tag *);
1495 void cxgbe_ratelimit_query(if_t, struct if_ratelimit_query_results *);
1496 #endif
1497 
1498 /* t4_filter.c */
1499 int get_filter_mode(struct adapter *, uint32_t *);
1500 int set_filter_mode(struct adapter *, uint32_t);
1501 int set_filter_mask(struct adapter *, uint32_t);
1502 int get_filter(struct adapter *, struct t4_filter *);
1503 int set_filter(struct adapter *, struct t4_filter *);
1504 int del_filter(struct adapter *, struct t4_filter *);
1505 int t4_filter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1506 int t4_hashfilter_ao_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1507 int t4_hashfilter_tcb_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1508 int t4_del_hashfilter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1509 void free_hftid_hash(struct tid_info *);
1510 
1511 static inline struct wrqe *
1512 alloc_wrqe(int wr_len, struct sge_wrq *wrq)
1513 {
1514 	int len = offsetof(struct wrqe, wr) + wr_len;
1515 	struct wrqe *wr;
1516 
1517 	wr = malloc(len, M_CXGBE, M_NOWAIT);
1518 	if (__predict_false(wr == NULL))
1519 		return (NULL);
1520 	wr->wr_len = wr_len;
1521 	wr->wrq = wrq;
1522 	return (wr);
1523 }
1524 
1525 static inline void *
1526 wrtod(struct wrqe *wr)
1527 {
1528 	return (&wr->wr[0]);
1529 }
1530 
1531 static inline void
1532 free_wrqe(struct wrqe *wr)
1533 {
1534 	free(wr, M_CXGBE);
1535 }
1536 
1537 static inline void
1538 t4_wrq_tx(struct adapter *sc, struct wrqe *wr)
1539 {
1540 	struct sge_wrq *wrq = wr->wrq;
1541 
1542 	TXQ_LOCK(wrq);
1543 	t4_wrq_tx_locked(sc, wrq, wr);
1544 	TXQ_UNLOCK(wrq);
1545 }
1546 
1547 static inline int
1548 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
1549     int len)
1550 {
1551 
1552 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
1553 }
1554 
1555 static inline int
1556 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
1557     const uint32_t *val, int len)
1558 {
1559 
1560 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
1561 }
1562 
1563 /* Number of len16 -> number of descriptors */
1564 static inline int
1565 tx_len16_to_desc(int len16)
1566 {
1567 
1568 	return (howmany(len16, EQ_ESIZE / 16));
1569 }
1570 #endif
1571